Not trusting the models's self-assessment is the right call, considering that the actual score summed up to 7.5 compared to the self-reported 6.5 :)